Cette annexe contient un article qui va paraître dans la revue EURASIP Journal on Wireless Communications and Networking, Special Issue on Ad Hoc Networks : Cross-Layer Issues. Il décrit un mécanisme de sélection de mode de transmission pour réseaux locaux sans fil 802.11 qui prend en compte les caractéristiques de l’application. Ce mécanisme a été présenté dans le chapitre 5.
102/ 116 D. ARTICLE MORSA
An Evaluation of Media-Oriented Rate Selection Algorithm for Multimedia Transmission in MANETs
Mohammad Hossein Manshaei, Thierry Turletti Thomas Guionnet
Plan`ete Project, INRIA Temics Project, IRISA-INRIA
2004 Route des Lucioles, BP-93 Campus de Beaulieu
06902 Sophia-Antipolis Cedex, France 35042 Rennes Cedex, France
E-mail:{manshaei,turletti}@sophia.inria.fr E-mail: Thomas.Guionnet@irisa.fr
Abstract— Current wireless LANs treat multimedia flows and
classical data flows alike. Typically, the same error control mechanisms are used for video flows which are generally tolerant but delay-sensitive, and TCP flows, which are error-intolerant and delay-insensitive. The performance of multimedia applications can be significantly improved by some degree of cross-layer awareness. In this paper, we focus on the optimization of real time multimedia transmission over 802.11 based ad-hoc networks. In particular, we propose a simple and efficient cross layer mechanism for dynamically selecting the transmission mode that considers both the channel conditions and characteristics of the media. This mechanism called MORSA (Media-Oriented Rate Selection Algorithm) targets loss-tolerant applications such as audio/video conferencing or VoD that do not require full reliable transmission. We provide an evaluation of this mechanism for MANETs using simulations with ns and analyze the video quality obtained with a fine grain scalable video encoder based on a motion-compensated spatio-temporal wavelet transform. Our results show that the proposed cross-layer approach achieves up to 4 Mbps increase in throughput and that the routing overhead decreases significantly. The transmission of a sample video flow over an 802.11a wireless channel has been evaluated with MORSA and compared with the traditional approach. Significant improvement is observed in throughput, latency and jitter while keeping a good level of video quality.
Index Terms— Ad hoc networks, Cross-Layer optimization,
IEEE 802.11 Wireless LAN, MANETs, Mode-selection algorithms.
I. INTRODUCTION
With recent performance advancements in computer and wireless communications technologies, mobile ad hoc networks (MANETs) are becoming an integral part of communication networks. The emerging widespread use of real-time voice, audio and video applications generates interesting transmission problems to solve over MANETs. Many factors can change the topology of MANETs such as the mobility of nodes or the changes of power level. For instance, power control done at the physical (PHY) layer can affect all other nodes in MANETs, by changing the levels of interference experienced by these nodes and the connectivity of the network, which impacts routing. Therefore, power control is not confined to the physical layer, and can affect the operation of higher level layers. This can be viewed as an opportunity for cross-layering design and
poses many new and significant challenges with respect to wired and traditional wireless networks. As soon as we want to optimize data transmission according to both the characteristics of the data and to the varying channel conditions, a cross-layering approach becomes necessary. Numerous cross layer protocols have already been proposed in the literature [1], [2], [3], [4], [5]. They focus on the interactions between the application, transport, network and link layers. With the recent interest on software radio designs [7], it becomes possible to make the PHY layer as flexible as the higher layers. Adaptive and cross layering interactions can now affect the whole stack of the communication protocol. Consequently, the classical OSI approach of providing a PHY layer as reliable as possible independently of the type of data transmitted becomes questionable.
In this paper, we focus on the optimization of real time multi-media transmission over 802.11 based MANETs. In particular, we propose a simple and efficient cross layer protocol which dynamically adjusts the transmission mode, i.e., the physical modulation, rate and possibly the Forward Error Correction (FEC). This protocol called MORSA (Media-Oriented Rate Selection Algorithm) is convenient for loss-tolerant (LT) ap-plications such as video or audio codecs that do not require 100% transmission reliability (i.e., a certain level of packet loss rate (PER) or bit error rate (BER) can be concealed at the receiver). Contrary to mail and file transfer applications, several multimedia applications, such as audio and video conferencing or video on demand (VoD) can tolerate some packet loss. For example, an MPEG video data flow can contain three different types of packet, Intra-Picture (I) frames, Prediction (P) frames and Bi-prediction (B) frames. I-frames are more important for the overall decoding of the video stream, because they serve as reference frames for P- and B-frames. Therefore, the loss of an I-frame has a more drastic impact on the quality of the video playback than the loss of other types of frames. In this respect, the frame loss requirement of I-frames is more stringent than those of P- and B-frames. Furthermore, as described in Section VI, some multimedia applications implement their own error control mechanisms [15] [16], making it inefficient to provide full reliability at the link layer.
D. ARTICLE MORSA 103/ 116
2
of the application and varying conditions of the channel. It se-lects the highest possible transmission rate while guaranteeing a specific bit error rate: the selected transmission mode varies with time depending on the PER or BER tolerance and on the signal-to-noise ratio (SNR) measured at the receiver. We show in this paper that by adaptively selecting the transmission mode according to both loss tolerance requirements of the application and varying channel conditions, the application-layer through-put can be significantly increased and more stability can be achieved in ad hoc routing. Finally, we evaluate the quality of a sample video transmitted over a wireless 802.11a channel using MORSA and compare it with the quality obtained when we do not take into account characteristics of the application (i.e. using the standard approach). Our results show that MORSA can reach a comparable video quality than the one obtained with the standard mechanism while using only a very low (5%) FEC overhead at the application level instead of the physical layer FEC (50% or 25%). This significantly decreases transmission delay of the application.
Throughout this paper, we assume that wireless stations use the Enhanced Distributed Channel Access (EDCA), proposed in the IEEE 802.11e [22] to support different levels of QoS. We have modified the NS simulation tool to evaluate the overall system efficiency when considering the interaction between layers in the protocol stack.
The rest of this paper is structured as follows. In Section II, we overview the salient features of the MAC and PHY layers in the 802.11 schemes. We also review some of the automatic rate selection algorithms that were proposed in the literature. In Section III we present related work about cross layer protocols in ad hoc networks. The MORSA scheme and a possible implementation within a 802.11-compliant device are discussed in Section IV. Simulation results with ns are analyzed in Section V. We evaluate quality of a sample video transmission over a wireless channel in Section VI. Finally, the conclusion is presented in Section VII.
II. BACKGROUND
Today, three different PHY layers are available for the IEEE 802.11 WLAN as shown in Table I.
The performance of a modulation scheme can be measured by its robustness against path loss, interferences and fading that causes variations in the received SNR. Such variations also cause variations in the BER, since the higher the SNR, the easier it is to demodulate and decode the received bits. Compared to other modulations schemes, BPSK has the mini-mum probability of bit error for a given SNR. For this reason, it is used as the basic mode for each PHY layer since it has the maximum coverage range among all transmission modes. As shown in Figure 1, each packet may be sent with two different rates [17]: its PLCP (Physical Layer Convergence Protocol) header is sent at the basic rate while the rest of the packet might be sent at a higher rate. The higher rate, used to transmit
the physical-layer payload, which includes the MAC header, is stored in the PLCP header.
Sent with the rate indicated in PLCP Mac Header + Payload Sent with Basic Rate
PLCP Header
Fig. 1. Data rates for packet transmission.
The receiver can verify that the PLCP header is correct (using CRC or Viterbi decoding with parity), and uses the transmission mode specified in the PLCP header to decode the MAC header and payload. The mode with the lowest rate is used to transmit the PLCP header. Transmission mode selection can be performed manually or automatically in each station. A number of rate selection algorithms have been proposed in the literature. They include the Auto Rate Fallback (ARF) [19], the Receiver-Based Auto Rate (RBAR) [18] and Miser [20] schemes. RBAR tries to select the best mode (i.e. the mode which the highest rate) based on the received SNR, while ARF uses a simple ACK-based mechanism to select the rate. MiSer is a protocol based on the 802.11a/h standards whose goal is to optimize the local power consumption. While all these automatic rate selection mechanisms try to adapt the transmission mode according to the channel conditions, we are not aware of any protocol that considers characteristics of the application.
Since MORSA is based on RBAR, we detail the latter here. In RBAR, the sender chooses a data rate based on some heuristic (e.g., the most recent rate that was used to successfully transmit a packet), and then stores the rate and the packet size into the Request to Send (RTS) control packet. Stations that receive the RTS can use the rate and packet size information to calculate the duration of the requested reservation. They update their Network Allocation Vectors (NAVs) to reflect the reservation. While receiving the RTS, the receiver uses the current channel state as an estimate of the channel state when the upcoming packet is supposed to be transmitted. The receiver then selects the appropriate rate with a simple threshold-based mechanism and includes this rate (along with the packet size) in a Clear to Send (CTS) control packet. Stations that overhear the CTS calculate the duration of the reservation and update their NAVs accordingly. Finally, the sender responds to the CTS by transmitting the data packet at the rate selected by the receiver. Note that nodes that cannot hear the CTS can update their NAVs when they overhear the actual data packet by decoding a part of the MAC header called the reservation subheader. Further information concerning RBAR, including implementation and performance issues in 802.11b is available in [18].
III. RELATEDWORK
Several cross layer mechanisms such as mechanisms for TCP over wireless links [1], [5], power control [6], medium access
104/ 116 D. ARTICLE MORSA
3
TABLE I
CHARACTERISTICS OF THE VARIOUS PHYSICAL LAYERS IN THEIEEE 802.11STANDARD.
Characteristic 802.11a 802.11b 802.11g
Frequency 5 GHz 2.4 GHz 2.4 GHz
Data Rates 6, 9, 12, 18, 24, 36, 48, 54 Mbps 1, 2, 5.5, 11 Mbps 1, 2, 5.5, 6, 9, 11, 12, 18, 22, 24, 33, 36, 48, 54 Mbps
Modulation BPSK, QPSK, 16 QAM, 64 QAM BPSK, QPSK, CCK BPSK, QPSK, 16 QAM, 64 QAM, CCK
FEC Rate 1/2, 2/3, 3/4 NA 1/2, 2/3, 3/4
Basic Transmission Mode BPSK, 6 Mbps, FEC 1/2 BPSK, 1 Mbps 802.11a (6 Mbps) or 802.11b (1 Mbps) basic modes
control [2], QoS providing [8], video streaming over wireless LANs [9], and deployment network access point [1] have been proposed.
The Mobileman European project [12] introduced inside the layered architecture the possibility that protocols belonging to different layers can cooperate by sharing network status information while still maintaining separation between the layers in protocol design. The authors propose applying triggers to the network status such that it can send signals between layers. In particular, This cross-layering approach addresses the security and cooperation, energy management, and quality-of-service issues.
The effect of such cross layer mechanisms on the routing protocol, the queuing discipline, the power control algorithm, and the medium access control layer performance have been studied in [2].
A cross-layer algorithm using MAC channel reservation control packets at the physical layer is described in [4]. This mechanism improves the network throughput significantly for mobile ad hoc networks because the nodes are able to perform an adaptive selection of a spectrally efficient transmission rate. [9] describes a cross-layer algorithm that employs different error control and adaptation mechanisms implemented on both application and MAC layers for robust transmission of video. These mechanisms are Media Access Control (MAC) retrans-mission strategy, application-layer Forward Error Correction (FEC), bandwidth-adaptive compression using scalable coding, and adaptive packetization strategies. Similarly a set of end-to-end application layer techniques for adaptive video streaming over wireless networks is proposed in [10]. In [11], the Adaptive Source Rate Control (ASRC) scheme is proposed to adjust the source rate based on the channel conditions, the transport buffer occupancy and the delay constraints. This cross layer scheme can work together with hybrid ARQ error control schemes to achieve efficient transmission of real-time video with low delay and high reliability. However, none of these algorithms have tried to adapt the physical layer transmission mode in 802.11 WLANs. More examples could be cited, but we are not aware of any cross layer algorithm that takes into account the physical layer parameters (e.g., PHY FEC) as explained in Section II.
It should be noted that standardization efforts are in progress to integrate various architectures. The important co-design of the physical, MAC and higher layers have been taken into
account in some of the latest standards like 3G standards (CDMA2000), BRAN HiperLAN2, and 3GPP (High Speed Downlink Packet Access) [1]. IEEE has also considered a cross-layer design in the study group on Mobile Broadband Wireless Access (MBWA).
IV. CROSSLAYERMODESELECTIONPROTOCOL This section describes the MORSA mechanism and dis-cusses implementation issues.
A. Algorithm Description
As we already mentioned, real-time multimedia applications can be characterized by their tolerance to a certain amount of packet loss or bit errors. These losses can be ignored (if they are barely noticeable by human viewers) or compensated at the receiver using various error concealment techniques. In our scheme, the sender is able to specify its loss tolerance (LT) such that the receiver uses both this information and the current channel conditions to select the appropriate transmission mode (i.e., rate, modulation, and FEC level). More precisely, the sender includes the LT information in each RTS packet to allow the receiver to select the best mode. The LT information is also included in the header of each data packet such that the receiver can decide whether or not to accept a packet. While receiving the RTS, the receiver uses the information concerning the channel conditions along with the information related to LT to select the best data rate for the corresponding packet. The selected rate is then transmitted along with the packet size in the CTS back to the sender, and the sender uses this rate to send its data packets. When a packet arrives at the receiver side, if the receiver is able to decode the PLCP header, it can identify the BER tolerance for the encoded payload. If the packet can tolerate some bit errors, it has to be accepted even if its payload contains errors. As will be shown later, our mechanism makes it possible to define new transmission modes that do not use FEC but that exhibit comparable throughput performance.
To take into account both the SNR and the LT informa-tion, we have modified the RBAR threshold1mechanism. For 802.11a, we assume that the receiver uses FEC Viterbi decod-ing. The upper bound on the probability of error provided in [13], [20] is used under the assumption of binary convolutional 1These thresholds are used to select the best transmission mode in the receiver.
D. ARTICLE MORSA 105/ 116
4
coding and hard-decision Viterbi decoding. Specifically, for a packet of length L (bytes) the probability of packet error can be bound by:
Pe(L) ≤ 1 − (1 − Pu)8L (1)
where the union bound Puof the first-event error probability is given by Pu= ∞ X d=df ree ad·Pd (2)
With df ree, the free distance of the convolutional code, ad the total number of error events of weight2 d and Pd, the probability that an incorrect path at distance d from the correct path is chosen by the Viterbi decoder. When hard decision decoding is applied, Pd is given by Equation 3, where ρ is the probability of bit error for the modulation selected in the physical layer3. Pd= Pd k=(d+1)/2 dk · ρk·(1 − ρ)d−k if d is odd 1 2· d d/2 ·ρd/2·(1 − ρ)d/2 if d is even +Pd k=d/2+1 dk · ρk·(1 − ρ)d−k (3) Figure 2 shows an example of the modifications made for the SNR threshold in RBAR with and without the media-oriented mechanism. Commonly, a BER at the physical layer smaller than10−5 is considered acceptable in wireless LAN applications. By using theoretical graphs of BER as function of the SNR for different transmission modes on a simple additive white Gaussian noise (AWGN) channel (see Figure 2), we can compute the minimum SNR values required. Now if a particular application can tolerate some bit errors (e.g. a BER up to the10−3as shown in Figure 2), the receiver can select the highest rate for the following data transmission corresponding to this SNR. For example in Figure 2, when the SNR is equal to5dB, the receiver can select a 9M bps data rate instead of a 6 M bps data rate if it is aware that the application can tolerate a BER less that10−3.
We have calculated the thresholds using Equations 1, 2, and 3 for an application that can tolerate up to 10−3 BER (See Table III in Section V.). The receiver can use arrays of thresholds that are pre-computed for different LTs.
In the following sections, we describe how such a mecha-nism can be implemented in 802.11-based WLANs. B. Implementation issues
We propose to implement MORSA with the help of the EDCA protocol [21], [23]. EDCA is one of the features that has been proposed by IEEE 802.11e to support QoS in WLANs
2We have used the adcoefficients provided in [14].
3In this paper we use Additive White Gaussian Noise (AWGN) channel model. BER = 0.00001 BER = 0.001 Change in Thresholds 1e−08 1e−07 1e−06 1e−05 0.0001 0.001 0.01 0 5 10 15 20 25 30 35 SNR (db) BPSK 6 Mbps BPSK 9 Mbps QPSK 12 Mbps QPSK 18 Mbps 16−QAM 36 Mbps 64−QAM 48 Mbps 64−QAM 54 Mbps BER 16−QAM 24 Mbps
Fig. 2. Bit error rate (BER) versus SNR for various transmission modes (802.11a).
[22]. In this protocol each QoS-enhanced station (QSTA) has 4 queues to support up to 8 User Priorities (UP). Figure 3 shows the QoS control field that is added to the MAC header in the 802.11e specification [22]. Bits 6 and 7 of this header can be used to indicate the loss tolerance information. Table II shows a possible meaning for these two bits in our media-oriented mechanism that should be defined in the process of connection setup. LT information is sent to the receiver by adding one byte to the RTS packets as illustrated in Figure 4.
Bit 0-3 Bit 4 Bit 5 Bit 6-7 Bit 8-15
Traffic ID Schedule Pending Ack Policy Reserved TXOP duration
Fig. 3. QoS control field in the 802.11e.
TABLE II LOSSTOLERANCE CLASSIFICATION.
Bit 6-7 Application Sensitivity
00 No tolerance in payload
01 Low loss tolerance in payload
10 Medium loss tolerance in payload 11 High loss tolerance in payload
Frame Control Length Rate & BYTES 2 2 6 6 1 4 FCS Tolerance Information Source Address Dest Address
Fig. 4. Modifications to the RTS header.
To make our mechanism operational, it is crucial to let the packets with corrupted payload reach the receiver’s applica-tion layer. As such, some modificaapplica-tions of the standard are